试析果树对靶喷雾系统中图像识别技术

《试析果树对靶喷雾系统中图像识别技术》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、中国农业大学硕士学位论文果树对靶喷雾系统中图像识别技术姓名：肖健申请学位级别：硕士专业：农业机械化工程指导教师：曾爱军;何雄奎20050601撼要针对果树羯虫害防治工作中劳动强庹大、防治器械落肢、农药污染严重镣问题，提出将机器视觉技术应用予果园对靶喷雾系统。通过对仿真粜树图像实时采集、分割获取树形和枝咔分布嵇息，并以十六避铡编码翔下位唤雾装置下达喷雾指导信息。对于每一帧图像(768X576象素)从采集茔Ⅱ下达喷雾信息的总处理时闯为0．579秒。本文重赢硪究果瓣髑像分割方法。罄先，采取基于颜色特征的努割。在筛选颜色特征时，突破前入定性评价颜色特征优劣的方法，引入以“

2、误判率”为指标的定量颜色特征评价嵛法。通邂“谖髑率”麴线可致嗡臻豹说骥蔷静颜甑特征在分割中静袭瑰，最终选定ExGl作为颜色特征。其次，以复杂环境中粜树图像分割为目的，初步探讨了基平纹理特征的图像分割方法。研究中将强像势残无蘩复浆纹懋送块(16x16象舞斡。蠹区块绞理转摄：能量、熵辅戳图像区块x彝Y搬标组成纹理隧块样本向蚩代表谈小区特征。对由±壤、果树枝叶、草地和天缴四种果闼中常见纹理会戚瓣实骏鏊鬣聚类，分类络莱歪确攀这t00N。来掬象素区块住鬟绩患，馁班糍豢秘烧撵为向攫特征时，这四种纹理合成图像分类派确率从4NN90％不等。表明在纹理分析中，纹理本身特授赣鞋绞壤区

3、块鑫黧信息哥疆提高分鲻效果。关键诲：暴辩瓣靶缕雾，匿豫谖羯，耩确藤芬，麟垂，纹瀵AbstractTheheavytellance011chemicaisraisemanyeuviromentalandeconomiccoucerns'thetriditonalsprayerand1)esitioideapplicationmethodcouldnotsuitthemordemorchardswaying．Inordertoreducechemicaluseandgetfarmerfreefromhardwork虹orchardspraying,§simulated

4、mJ-timemachinevisionbasedroboticorchardsprayingsystemwasbuflded，thatincludethreeparts：(1)leal*娃矬le啦eimagecat吐ingsyslcm；(2)treeimagepreprocossingandrecognitionsystem；《3)imageexpressionandmessagecommunicationsystem．Ittook0．579sfortheholesystemtoprocessingODeimage(768>(576pixels)，represe

5、ntinga3mby3mregionofoneu'ee．TheemphasisofthisdissertationWaslaidonthestudyoftreeimagerecognition．Firstly,colorWaSusedtodistinguishtreeCrOWfromthebackground．Forsakeoffindingtheproperfeatureincolormodel，ultimate-measurementlc∞mcy《l孺弧}wasintroducedtodecidewhichfeature醐糖p∞舜a溆Accordingthis

6、evaluationmethod,Ex01瑚骷selectedasthecolorfeatureinUeeimagesegmentation．Secondly，texturalfeatureanalysiswasusedtoidentifyfleeclrown．Theimage(768X576plxds)wasdividedintorectangles(16X16pixels)．EnergyandentropyderivingfromGLCMandcoordinate(x,y)ofeachmctangIeWereselectedasindexofvectorwhi

7、chrepresenteveryrectangles．Theman-madeimagewhichincludefourdifereuttexture(soil，leafsoffruit舡ee，grassplotandsky)aretested,theresultshowedthateachkindoftexturerectangelWaSrecognisedwith100％,Butifthevectorofmctangelwithoutcoordina垂e,theac：cl】tratenessofrecognitionWasdifferent(from毒％to粥％

8、XSoth